Anthropic leads five of six domains in the summer safety index

The Future of Life Institute's Summer 2026 index grades nine companies on 37 indicators across six domains. Anthropic takes the highest overall grade. The more useful reading is what the index can measure, which is mostly disclosure.

The Future of Life Institute's Summer 2026 AI Safety Index evaluates nine leading companies on 37 indicators across six domains. Anthropic again earns the highest overall grade and leads five of the six, on the strength of transparency, an established safety framework, technical research and governance.

What an index like this can and cannot see

Every one of those four strengths is a disclosure. A published framework, documented governance, papers in the open. They are real and they are not nothing — a company that will not say what its policy is has no policy anyone can rely on.

But an index built from public artefacts rewards publishing. A lab doing excellent unpublished safety work scores badly; a lab publishing extensively while shipping aggressively scores well. That is not a flaw to be fixed so much as a limit to be stated, and it should be stated every time the ranking is cited.

Nine companies, thirty-seven indicators

The breadth is the useful part. Six domains across nine firms makes the shape of the field visible — where everyone is weak, where one firm is carrying a category alone. A single overall grade compresses exactly that information away, which is why the domain-level detail is worth more than the headline it generates.

Alongside it

Thirty-five of 223 oral presentations at ICLR 2026 were safety-related, spanning alignment, jailbreaks, interpretability, bias, privacy, hallucination and watermarking. Roughly one paper in six at a top venue is now a safety paper — a better measure of where research attention actually sits than any index of company behaviour.

Future of Life Institute — AI Safety Index — Summer 2026 → · ICLR 2026 review — 35 AI-safety oral papers at ICLR 2026 → · Cloud Security Alliance — The Alignment Gap: Control Failure Risk Before ASI → · AI Security Institute — Research & Publications →